Shai Avidan (Tel Aviv University)- Enhanced 3D Representation and Alignment
Abstract: 3D Representation enjoyed a resurgence with the introduction of NeRF and Gaussian Splatting. In this talk, I will describe three projects aimed at improving our abilities to work with them. The first, VF-NeRF, is a novel method that directly aligns two NeRF models, without going through some intermediate representation. Next, I will describe a new method for cleaning NeRF from “floaters” by enforcing a new prior that we term “Free Space Prior”. Lastly, I will shift my attention to Gaussian Splatting and describe a novel way to decompose the Gaussians into groups that correspond to different frequency bands in the input images. This underpins various applications such as Level-Of-Details representation, Streaming, and artistic modeling, to name a few.
Speakers
Shai Avidan
Shai Avidan is a Professor in the School of Electrical Engineering at Tel Aviv University. He is widely recognized for his pioneering contributions to computer vision, image processing, and computational photography. Shai Avidan received his Ph.D. degree from the School of Computer Science, Hebrew University, Jerusalem, Israel, in 1999. He is currently a professor in the Faculty of Engineering, Tel Aviv University. He worked for Adobe, Mitsubishi Electric Research Labs, MobilEye, and Microsoft Research. He has published extensively in the fields of object tracking in video and 3D object modeling from images. He is interested in everything that involves pixels.
